1. Field of the Invention
This invention relates generally to a current perpendicular to plane (CPP) magnetic random access memory (CPP-MRAM) cell formed using a magnetic tunneling junction (MTJ) as the basic memory element, wherein a spin torque transfer (STT) effect is used to change the magnetization direction of the MTJ ferromagnetic free layer.
2. Description of the Related Art
The conventional magnetic tunneling junction (MTJ) device is a form of ultra-high magnetoresistive device in which the relative orientation of the magnetic moments of parallel, vertically separated, upper and lower magnetized layers controls the flow of spin-polarized electrons tunneling through a very thin dielectric layer (the tunneling barrier layer) formed between those layers. When injected electrons pass through the upper layer they are spin polarized by interaction with the magnetic moment of that layer. The majority of the electrons emerge polarized in the direction of the magnetic moment of the upper layer, the minority being polarized opposite to that direction. The probability of such a polarized electron then tunneling through the intervening tunneling barrier layer into the lower layer then depends on the availability of states within the lower layer that the tunneling electron can occupy. This number, in turn, depends on the magnetization direction of the lower electrode. The tunneling probability is thereby spin dependent and the magnitude of the current (tunneling probability times number of electrons impinging on the barrier layer) depends upon the relative orientation of the magnetizations of magnetic layers above and below the barrier layer. The MTJ device can therefore be viewed as a kind of multi-state resistor, since different relative orientations (e.g. parallel and antiparallel) of the magnetic moments will change the magnitude of a current passing through the device. In a common type of device configuration (spin filter), one of the magnetic layers has its magnetic moment fixed in direction (pinned) by exchange coupling to an antiferromagnetic (AFM) layer, while the other magnetic layer has its magnetic moment free to move (the free layer). The magnetic moment of the free layer is then made to switch its direction from being parallel to that of the pinned layer, whereupon the tunneling current is large, to being antiparallel to the pinned layer, whereupon the tunneling current is small. Thus, the device is effectively a two-state resistor. The switching of the free layer moment direction (writing) is accomplished by external magnetic fields that are the result of currents passing through conducting lines adjacent to the cell. Once the cell has been written upon, the circuitry must be able to detect whether the cell is in its high or low resistance state, which is called the “read” process. This process must both measure the resistance of the written-upon cell and then compare that resistance to that of a reference cell in a fixed resistance state, to determine if the written-upon cell is in its high or low state. Needless to say, this process also introduces some statistical difficulties associated with the variation of resistances of the cells.
FIG. 1 is a highly schematic drawing showing an overhead view of a conventional MRAM cell comprising an MTJ cell element (1000) positioned between (or at the intersection of) vertically separated orthogonal word (200) and bit (100) lines. The cell element (1000) is drawn with a slightly elliptical horizontal cross-section because such an anisotropic shape (“shape anisotropy”) produces a corresponding magnetic anisotropy within the free layer that assists its magnetic moment in retaining a thermally stable fixed position after switching fields have been turned off. The direction along the free layer in which it is energetically favorable for the moment to remain and from which it should be difficult to switch the magnetic moment unintentionally (as with thermal effects), the longer direction in this case, is called the “easy axis” of the layer. The axis perpendicular to the easy axis is called the “hard axis.” The fields produced by currents in each of the two lines are between about 30 to 60 Oersteds in magnitude. According to the diagram, the word line field will be along the easy axis of the cell element, the bit line field will be along the easy axis.
The use of magnetic fields externally generated by current carrying lines (as in FIG. 1) to switch the magnetic moment directions becomes problematic as the size of the MRAM cells decreases and, along with their decrease, so must the width of the current carrying lines. The smaller width lines require greater currents to produce the necessary switching fields, greatly increasing power consumption.
For this reason, a new type of magnetic device, called a spin transfer device, described by Slonczewski, (U.S. Pat. No. 5,695,864) and Covington (U.S. Pat. No. 7,006,375), has been developed, that seems to eliminate some of the problems associated with the excessive power consumption necessitated by external switching fields. The spin transfer device shares some of the operational features of the conventional MTJ cell (particularly the read process) described above, except that the switching of the free layer magnetic moment (the write process) is produced by passage of the spin polarized current itself. In this device, unpolarized conduction electrons passing through a first magnetic layer having its magnetic moment oriented in a given direction (such as the pinned layer) are preferentially polarized by their passage through that layer by a quantum mechanical exchange interaction with the polarized bound electrons in the layer. Such a polarization can occur to conduction electrons that reflect from the surface of the magnetized layer as well as to those that pass through it. The efficiency of such a polarization process depends upon the crystalline structure of the layer. When such a stream of polarized conduction electrons subsequently pass through a second magnetic layer (such as the free layer) whose polarization direction is not fixed in space, the polarized conduction electrons exert a torque on the bound electrons in the magnetic layers which, if sufficient, can reverse the polarization of the bound electrons and, thereby, reverse the magnetic moment of the magnetic layer. The physical explanation of such a torque-induced reversal is complicated and depends upon induction of spin precession and certain magnetic damping effects (Gilbert damping) within the magnetic layer (see Slonczewski, below). If the magnetic moment of the layer is directed along its easy magnetic axis, the required torque is minimized and the moment reversal occurs most easily. The use of a current internal to the cell to cause the magnetic moment reversal requires much smaller currents than those required to produce an external magnetic field from adjacent current carrying lines to produce the moment switching. Much recent experimental data confirm magnetic moment transfer as a source of magnetic excitation and, subsequently, magnetic moment switching. These experiments confirm earlier theoretical predictions (J. C. Slonczewski, J. Magn. Mater. 159 (1996) LI, and J. Z. Sun, Phys. Rev. B., Vol. 62 (2000) 570). These latter papers show that the net torque, Γ, on the magnetization of a free magnetic layer produced by spin-transfer from a spin-polarized DC current is proportional to:Γ=snmx(nsxnm),  (1)Where s is the spin-angular momentum deposition rate, ns is a unit vector whose direction is that of the initial spin direction of the current and nm is a unit vector whose direction is that of the free layer magnetization and x symbolizes a vector cross product. According equation (1), the torque is maximum when ns is orthogonal to nm.
Referring to FIG. 2, there is shown a schematic illustration of an exemplary related art MTJ cell element (such as that in FIG. 1) being contacted from above by a bit line (100) and from below by a bottom electrode (300). The bottom electrode is in electrical contact, through a conducting via (80), with a CMOS transistor (500) that provides current to the MTJ element when the element is selected in a read or write operation.
Moving vertically upward from bottom electrode to bit line this prior art storage device consists of an underlayer (1), which could be a seed layer or buffer layer, an antiferromagnetic pinning layer (2), a synthetic antiferromagnetic (SyAF) pinned reference layer (345), consisting of a first ferromagnetic layer (3), a non-magnetic spacer layer (4) and a second ferromagnetic layer (5), a non-conducting tunneling barrier layer (6), a ferromagnetic free layer (7) and a non-magnetic capping layer (8). Arrows, (20) and (30), indicate the antiparallel magnetization directions of the two ferromagnetic layers (3) and (5) of the SyAF pinned layer (345). The double-headed arrow (40) in free layer (7) indicates that this layer is free to have its magnetic moment directed in either of two directions.
Referring again to FIG. 2 it is noted that when a certain critical current (arrow (50) is directed from bottom to top (layer (1) to layer (8)), the free layer magnetization (40) would be switched to be opposite to the direction of the reference layer's magnetization (30) by the spin-transfer torque. This puts the MTJ cell into its high resistance state.
Conversely, if the current is directed from top to bottom (60), the free layer magnetization (40) would be switched, by torque transfer of angular momentum, to the same direction as that of the pinned reference layer's direction (30), since the conduction electrons have passed through that layer before entering the free layer. The MTJ element is then in its low resistance state.
Referring again to FIG. 2, this entire configuration represents a schematic diagram of a single spin-RAM memory cell that utilizes the spin transfer effect (denoted hereinafter as an STT-RAM) for switching an MTJ type element. In this paper, we will use the term “element” to describe the basic MTJ structure comprising a tunneling barrier layer sandwiched between ferromagnetic fixed and free layers. We shall use the term “memory cell” to denote the combination of the MTJ element incorporated within the circuitry shown that permits the element to be written on and read from. Such circuitry includes intersecting current carrying lines that allow a particular element to be accessed and also a CMOS transistor that allows a current to be injected into the element. The word line provides the bit selection (i.e., selects the particular cell which will be switched by means of a current passing through it between the bit line and the source line) and the transistor provides the current necessary for switching the MTJ free layer of the selected cell. Although it is not shown in this simplified figure, the cell is read by applying a bias voltage between the bit line and source line, thereby measuring its resistance and comparing that resistance with a standard cell in the circuit (not shown).
The critical current for spin transfer switching, Ic, is generally a few milliamperes for an 180 nm sub-micron MTJ cell (of cross-sectional area A approximately A=200 nm×400 nm). The corresponding critical current density, Jc, which is Ic/A, is on the order of several 107 Amperes/cm2. This high current density, which is required to induce the spin transfer effect, could destroy the insulating tunneling barrier in the MTJ cell, such as a layer of AlOx, MgO, etc.
The difference between an STT-RAM and a conventional MRAM is only in the write operation mechanism; the read operation is the same for both types of cell. In order for the spin transfer magnetization mechanism switching to be viable in the 90 nm MTJ cell structure and smaller, the critical current density must be lower than 106 A/cm2 if it is to be driven by a CMOS transistor that can typically deliver 100 μA per 100 nm of gate width. For STT-RAM applications, the ultra-small MTJ cells must exhibit a high tunnel magnetoresistance ratio, TMR=dR/R, much higher than the conventional MRAM-MTJ that uses AlOx as a tunneling barrier layer and has a NiFe free layer. Such MRAM-MTJ cells have a dR/R˜40%. It has recently been demonstrated (S. C. Oh et al., “Excellent scalability and switching characteristics in Spin-transfer torque MRAM” IEDM2006 288.1 and “Magnetic and electrical properties of magnetic tunnel junction with radical oxidized MgO barriers,” IEEE Trans. Magn. P 2642 (2006)) that a highly oriented (001) CoFe(B)/MgO/CoFe(B) MTJ cell is capable of delivering dR/R>200%. Furthermore, in order to have a satisfactory “read margin”, TMR/(Rpcovariance), where Rp is the MTJ resistance for parallel alignment of the free and pinned layers, of at least 15 and preferably >20 is required. It is therefore essential to find a method of fabricating the CoFe(B)/MgO/CoFe(B) MTJ cell with a good read margin for read operation. Note, “Rpcovariance” indicates the statistical spread of Rp values.
In MRAM MTJ technology, Rp is as defined above and Rap is the MTJ resistance when the free and pinned layers have their magnetizations aligned in an antiparallel configuration. Uniformity of the TMR ratio and the absolute resistance of the cell are critical to the success of MRAM architecture since the absolute value of the MTJ resistance is compared to the resistance of a reference cell during the read operation. If the active device resistances in a block of memory show a high variation in resistance (i.e. high Rp covariance, or Rap covariance), a signal error can occur when they are compared with the reference cell. In order to have a good read margin TMR/(Rpcovariance), should have a minimum value of 12 and most preferably be >20.
To apply spin transfer switching to the STT-RAM, we have to decrease Ic by more than an order of magnitude. The intrinsic critical current density, Jc, is given by Slonczewski (J. C. Slonczewski, J. Magn. Mater. 159 (1996) LI,) as:Jc=2eαMstF(Ha+Hk+2πMs)/hη  (1)where e is the electron charge, α is the Gilbert damping constant, tF is the free layer thickness, h is the reduced Planck's constant, η is the spin-transfer efficiency (related to the tunneling spin polarization factor of the incident spin-polarized current), Ha is the external applied field, Hk is the uniaxial anisotropy field and 2πMs is the demagnetization field of the free layer. Normally the demagnetization field is much larger than the two other magnetic fields, so equation (1) can be rewritten:Ic˜αMsV/hη,  (2)where V is the magnetic volume, V=MstFA, which is related to the thermal stability function term, KuV/kbT, which governs the stability of the magnetization relative to thermally-induced fluctuations.
M. Hosami (“A novel nonvolatile memory with spin torque transfer magnetization switching: Spin RAM” 2005 IEDM, paper 19-1), discusses a Spin-RAM 4 Kbit array which is fabricated using a stack of the following form: CoFeB/RF-sputtered MgO/CoFeB with a MnPt pinning layer. This MTJ stack is processed using 350° C., 10 KOe annealing. The cell size is a 100 nm×150 nm oval. Patterning of such sub 100 nm oval MTJ elements is done using e-beam lithography. The tunnel barrier layer is (001) crystallized MgO whose thickness is less than 10 angstroms for the desired RA of about 20 Ω-μm2. Intrinsic dR/R of the MTJ is 160%, although under operational conditions (0.1 V bias, for read determination) it is about 100%. Using a current pulse width of 10 ns, the critical current density is about 2.5×106 A/cm2. The amounts to a critical current of 375 μA. The distribution of write voltages for the array, for the high resistance state to the low resistance state has shown a good write margin. Resistance distributions for the high and low resistance states has a sigma (Rp covariance) around 4%. Thus, under operational conditions, (TMR/Rp covariance) is 25. This is equivalent to the conventional 4-Mbit CoFeB/AlOx/NiFe MTJ-MRAM in which dR/R (0.3V biased) typically is about 20-25%. For a Rp covariance=1%, TMR/(Rp covariance) is >20.
S. C. Oh et al., cited above, describes an STT-RAM utilizing spin torque transfer where a CoFeB/RF-sputtered MgO/CoFeB was processed with a 360° C.-10 KOe annealing. Pinning layer for the stack is MnPt. MgO thickness is controlled to less than 10 angstroms to give an RA of about 50 Ω-μm2. MRAM circuits made of sub-100 nm MTJ cells were made using conventional deep UV lithography. For the 80 nm×160 nm MRAM MTJ, Jc at 10 ns pulses is about 2.0×106 A/cm2. TMR at 400 mV bias is about 58% and the read margin, TMR/Rp covariance, for 100 nm×200 nm cells is 7.5, corresponding to a Rp covariance of 7.8%.
J. Hayakawa et al. (“Current—driven magnetization switching in CoFeB/MgO/CoFeB magnetic tunnel junctions, Japn. J. Appl. Phys. V 44, p. 1267 (2005)) has reported critical current densities of 7.8 and 8.8×105 and 2.5×106 with 10 ns pulse width, for MTJ cells processed with 270, 300 and 350° C. annealing. MgO barrier layer is about 8.5 angstroms thick, yielding a RA of about 10 Ω-μm2. Intrinsic MR as a function of the annealing temperature for an MTJ stack formed of Co40Fe40B20/MgO/Co40Fe40B20 with a 20 angstrom thick Co40Fe40B20 free layer are 49, 73 and 110% respectively. It is noted that the free layer in an MTJ processed at 270 and 300° annealing temperatures is amorphous. The pinning layer for the MTJ stack was IrMn.
Y. Huai et al. (“Spin transfer switching current reduction in magnetic tunnel junction based dual spin filter structures” Appl. Phys. Lett. V 87, p 222510 (2005)) have reported on spin-transfer magnetization transfer of a dual spin valve of the following configuration:Ta/MnPt/CoFe/Ru/CoFeB/Al2O3/CoFeB/Spacer/CoFe/MnPt/TaIt is noted that the free layer of the dual structure is made of a low saturation moment (approx. 1000 emu/cm3) amorphous CoFeB. The nominal MTJ size is 90 nm×140 nm. RA is about 20 Ω-μm2 and dR/R is about 20%. For a dual spin-filter (DSF) structure, the free layer experiences the spin transfer effect on both faces, so the critical current density has been reduced to approx. 1.0×106 A/cm2.
C. Horng et al. (docket No. HMG06-042. “A novel MTJ to reduce spin-transfer magnetization switching current”) is assigned to the same assignee (Magic Technologies) as the present invention and fully incorporated herein by reference. Horng et al. have produced an STT-MTJ test structure that includes a MTJ stack of the form:Ta/NiCr/MnPt/Co75Fe25/Ru7.5/Co60Fe20B20—Co75Fe25/(NOX)MgO11/Co60Fe20B20/TaWhich is processed at 265° C.-2 hrs-10 KOe annealing, so that the Co60Fe20B20 remains amorphous. It is noted that the pinning layer is MnPt. RA of the MTJ is controlled to less than 10 Ω-μm2 and intrinsic dR/R is about 100%. For the 100 nm×150 nm size MTJ, patterned using conventional photo-lithography of the 180 nm node technology, dR/R at 0.1 V bias is about 70-80%. Due to the fact that no array was constructed, there was no determination of Rp covariance. However, the covariance for a conventional MRAM of the same basic MTJ structure, but 200 nm×325 nm was measured to about 3.5%. Extrapolation to the 100 nm×150 nm size of the STT-MTJ predicts that the covariance would be about 7%. This value would not be sufficient to provide a good read margin.
The TMR sensor currently under production at Headway Technologies uses an MTJ element of the form:Ta/Ru/IrMn/CoFe/Ru/CoFeB/CoFe/MgO/CoFe—NiFe/NiFeHfIn this configuration, the pinning layer is IrMn. TMR sensor size when the resistance measurements are made (i.e., unlapped) is 100 nm×500 nm. Patterning is done using conventional photo-lithography of the 180 nm node technology. Rp covariance across the 6″ wafer for that sensor size is about 3%. Scaling to a 100 nm×150 nm size, the covariance is projected to be about 5%.
It should be noted that to obtain improvement in Rp covariance, the photo-lithography using the 65-90 nm node technology, as is now practiced in semiconductor technology, would be viable.
C. Bilzer et al. (“Study of the dynamic magnetic properties of soft CoFeB films”, J. Appl. Phys. V 100, 053903 (2006)) has measured the magnetization damping parameters for the ion-beam deposited Co72Fe18B10 film as a function of film thickness and crystalline state. Amorphous Co72Fe18B10 showed low damping with α between 0.006 and 0.008, which was thickness independent. Crystalline Co80Fe20 shows a damping factor that is approximately a factor of 2 higher.
M. Oogane et al. (“Magnetic damping in ferromagnetic thin film”, Japn. J. Appl. Phys. V 45, p 3889 (2006)) have measured the Gilbert damping factor for the ternary Fe—Co—Ni and CoFeB films. As shown in FIG. 6, a low damping constant is measured for the Fe rich FeCo and the Fe—Ni binary alloys. For the CoFeB alloys, as shown in FIG. 7, the damping constant is 0.0038 and 0.010 respectively for amorphous Co40Fe40B20 and Co60Fe20B20.
The above prior art tends to imply that:
1. Jc is greater than 2×106 A/cm2 for the CoFeB/MgO/CoFeB MTJ with a crystalline CoFeB free layer.
2. Jc less than 1.0×106 A/cm2 is achievable for the CoFeB/MgO/CoFeB MTJ with an amorphous CoFeB free layer.
3. The Rp covariance for an STT-RAM MTJ made, using conventional 180 nm node photo-lithography, with an MnPt (pinning)/CoFe(B)MgO/CoFeB structure is greater than or equal to 7.5%, while a covariance that is less than or equal to 5% may be achievable for the MTJ with an IrMn pinning layer.4. A low damping factor free layer is critical for reducing the spin-torque magnetization switching current.
An examination of the patented prior art shows an increasing number of inventions utilizing the STT approach to MRAM switching. Although this prior art describes many different MTJ stack configurations and layer materials, none of them address the particular combination of conclusions that we have drawn and that are listed above in 1. through 4.
Shimazawa et al. (US Patent Application 2007/0086120), Ashida et al. (US Patent Application 2007/0076469) and Huai et al. (US Patent Application 2006/01022969) all teach an AFM layer comprising IrMn.
Nguyen et al. (U.S. Pat. No. 6,958,927) and Huai et al. (U.S. Pat. No. 7,126,202) teach that a first AFM layer is preferably IrMn.
Huai et al. (U.S. Pat. No. 6,967,863) discloses that an AFM layer is preferably IrMn or PtMn.
Huai et al. (U.S. Pat. No. 7,106,624) states that the AFM is preferably PtMn but “nothing prevents” the use of IrMn instead.
Covington (U.S. Pat. No. 7,006,375) shows a pinned layer that can be either IrMn or PtMn.
Pakala et al. (US Patent Application 2006/0128038) discloses that seed layers may be used to provide a desired texture to the AFM layer. For example, if IrMn is used as the AFM layer, then a TaN layer should be used.
The present invention will describe a spin transfer MRAM device in which a new form of free layer, combined with an IrMn pinning layer will address the issues raised above in statements 1-4.